More helper functions

Nobody has claimed this yet.

I think the code in partools could be made substantially more elegant with a few helper functions:

``` R
clusterRequireNamespace <- function(cl, pkg) {
clusterCall(cl, requireNamespace, package = pkg, quietly = TRUE)
}
clusterRequirePackage <- function(cl, pkg) {
clusterCall(cl, library, package = pkg, character.only = TRUE)
}
clusterEval < function(cl, expr) {
clusterCall(cl, eval, expr, env = .GlobalEnv)
}

clusterAssign <- function(cl, name, value) {
clusterCall(cl, assign, x = name, value = value)
}

asCall <- function(x) {
if (inherits(x, "formula")) {
stopifnot(length(x) == 2)
x[[2]]
} else if (is.atomic(x) || is.name(x) || is.call(x)) {
x
} else {
stop("Unknown input")
}
}

clusterEvalI <- function(cl, expr, ...) {
args <- lapply(list(...), asCall)
interpolated <- methods::substituteDirect(asCall(expr), args)

clusterEval(cl, interpolated)
}
```

This allows (e.g.) `fileread` to be rewritten as:

``` R

fileread <- function (cls, fname, dname, ndigs, header = FALSE, sep = " ",
usefread = FALSE) {
if (usefread) {
clusterRequireNamespace(cls, "data.table")
clusterEvalI(cls, ~ myfread <- data.table::fread))
} else {
clusterEvalI(cls, ~ myfread <- read.table))
}

clusterEvalI(cl,
~ `_mychunk` <- filechunkname(fname, ndigs),
fname = fname, ndig
s = ndgs
)

clusterEvalI(cls,
~ dname <- myfread(`_mychunk`, header = header, sep = sep),
dname = as.name(dname),
header = header,
sep = sep
)
}
```

which I think is rather more clear since it focusses on the computation, rather than the details of the call manipulation (and working with the AST is obviously more elegant than working with strings)

(Also note the use of `_mychunk` as a variable name - using a non syntactic variable should decrease the chances of accidentally clobbering an object that the user created)

I haven't tested this code yet, but I'm happy to do so if you're interested in the idea.

The issue proposes helper functions for cluster calls and AST substitution, including a rewrite of fileread, but names no files or tests. Start by locating the existing fileread and related call-manipulation code, then assess the proposed helpers and test the behavior. Done means the helpers are implemented and the fileread example works without regressions.

Written by the indexing model from the issue text.